A survey of multi-label classification based on supervised and semi-supervised learning

Multi-label classification algorithms based on supervised learning use all the labeled data to train classifiers. However, in real life, many of the data are unlabeled, and it is costly to label all the data needed. Multi-label classification algorithms based on semi-supervised learning can use both labeled and unlabeled data to train classifiers, resulting in better-performing models. In this paper, we first review supervised learning classification algorithms in terms of label non-correlation and label correlation and semi-supervised learning classification algorithms in terms of inductive methods and transductive methods. After that, multi-label classification algorithms are introduced from the application areas of image, text, music and video. Subsequently, evaluation metrics and datasets are briefly introduced. Finally, research directions in complex concept drift, label complex correlation, feature selection and class imbalance are presented.